A NNALI DELLA
S CUOLA N ORMALE S UPERIORE DI P ISA Classe di Scienze
N EIL S. T RUDINGER
Remarks concerning the conformal deformation of riemannian structures on compact manifolds
Annali della Scuola Normale Superiore di Pisa, Classe di Scienze 3
esérie, tome 22, n
o2 (1968), p. 265-274
<
http://www.numdam.org/item?id=ASNSP_1968_3_22_2_265_0>
© Scuola Normale Superiore, Pisa, 1968, tous droits réservés.
L’accès aux archives de la revue « Annali della Scuola Normale Superiore di Pisa, Classe di Scienze » (
http://www.sns.it/it/edizioni/riviste/annaliscienze/) implique l’accord avec les conditions générales d’utilisation (
http://www.numdam.org/conditions). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une infraction pénale.
Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
REMARKS CONCERNING THE CONFORMAL
DEFORMATION OF RIEMANNIAN
STRUCTURES ON COMPACT MANIFOLDS
NEIL S. TRUDINGER
(*)
§
1Introdnction.
Yamabe
[8] seeks
to establish thefollowing
result:THEOREM :
Any compact
C°° Riemannianmanifold of
dimension ~a > 3can be
deformned conformally
to a C°° Riemannian structureof
constant scalar curvature.The author has
recently
noticed that theproof
in[8]
appears incom-plete, causing
thevalidity
of Yamabe’s theorem to be in doubt. The pur- pose of this paper,apart
fromindicating
this error, is to establish the re-sult itself
under, however,
some restriction on the curvature of the manifold(Section 3)
and also to prove arelated regularity
result forarbitrary manifolds, naniely
that weak solutions of the deformationproblem
arenecessarily
smooth ones
(Section 4).
The
partial
results of Section 3 are sufficient to showthat,
in morethan 3
dimensions,
a structure of constantpositive
scalar curvature may betopologically
deformed into one of constantnegative
scalar curvature.This had
originally
been derivedby
Aubin[1]
from Yamabe’s theorem. Ho-pefully,
aproof
of Yamabe’s theorem forarbitrary compact
manifolds will be found.We let be au n
dimensional,
C°° Riemannianmanifold, n ;:>
3 and let gij denote its fundamentalpositive
definite tensor. We consider anarbitrary
Pervennto alla Rodazione il 28 Nov. 1967.
(*) This research was partially supported by a Ford Foundation Po8tdoctoral grant at the Conrant Institute of Mathematical Scienoed, New York University.
conformal deformation of the
writing
it aswhere u
> 0,
E C°°(M).
_Let
-By R
denote the scalar curvatures of the g,jre8pectively.
Thenaccording
to[8],
we have theequation
where
denotes the
Laplace-Beltrami
ioperator corresponding
toThus the
problem
ofconformally deforming
the structure on M to oneof constant scalar curvature is
equivalent
toproving
the existence of a con-stant R and
positive,
C°°(M)
function usatisfying equation (2).
We take up thestudy
of thisequation
in thefollowing
sections.The author is
grateful
to R. Gardner and J. Moser for many useful discussionsconcerning
thisproblem.
§
2Yamabe’s approach.
Yamabe
[8] considers,
instead ofequation (2),
theequations
proving
THEOREM 1.
(Yamabe)
For any qN,
there exists a con.3tant~~q
and apositive
C°°function
uq(normalized by f iuq Bq d v = ~ ) ( 1 ) satisfying equation (3).
M
(1) We denote the volume eletueut on M by dv and assume
267
The solutions
Âq,
7 ttq are characterizedby
thevariational problem
ofminimizing
the functionalover the class of’ functions in
W2 (2) satisfying
We denote the above class
by Aq
so thatIt is
readily
seenthat 1
isnon-increasing in q
and that as q --~~~V, iq
2013~ whereSince the
proof
in[8]
of Theorem 1(called
Theorem Bthere)
is some-what
unnecessarily involved,
we insert here asimpler,
more direct version.WTe let ,u
>
0 denote theellipticity
constant of A u so that for allpoints
of J.1f.The
proof
issplit
into twostages :
(i)
Existence in WTe choose a« minimizing
sequence », ie a se- quence E~4~
with as n --~ oo. Sinceurn)
is bounded in(m),
(2) The Sobolev spaces for positive integer and are defined by
and
where a~~ is a mnMi-index and is to be understood in the seiiee of distri- bution theory.
it is
weakly precompact.
We show that is in factprecompact
inWl
The
argument
is a standard one in the calculus of variations(cf. [6]). By
the Sobolev
imbedding theorem,
asubsequence
of theu(nJ, (which
we im-mediately
relabel as 11(n)itself)
converges in to afunction uq
inA,y.
Hence for
arbitrary
E>
0 andsufficiently large
m, ndepending
on tThus
for
sufficiently large
m, n. It then follows(by (7))
that a«?i) converges to zcq inlV21
We thenclearly
haveF (u,)
and hence Uq is a solution of the variationalproblem.
Note that we may choose the > 0 so thata.e. Also the
vanishing
of the first variationyields
the Euler-La-grange
equations
for ’Uq, iefor
all ~
E1111 (M)
The function 1lq is thns a weak solution ofequation (.3).
(ii)
Sntoothne8s aoidpositivety.
Wequote
a result fromelliptic theory.
LEMMA. Let u be a tweak
(1~1 )), non-negative,
solution in Moj’
thelinear
equation
. where
f
ELr (31),
r> n/2.
Then u ispositive
and bounded and we the estimates. The lemma may be
proved
in various ways. Yamabe[8]
proves and usessimilar statements. These are also consequences of Moser’s work
(see [3],
f5],
or[7]).
269
To
apply
the lemma to thepresent situation,
we observe that u sati-sfies a linear
equation
of form(9)
withwhere
Thus zcq is bounded and
positive. Elliptic regularity theory [2]
then guaran- tees that uq is C°°.To establish Yamabe’s theorem from Theorem 1 it suffices to show that the functions
uq
areequibounded
and henceequicontinuous.
Suchis the purpose of the second
part
of Yamabe’s work([8],
Theorem(‘).
Inhis
proof, however,
theinequality (6.2)
appears to be in error. The correct form should be theinequality
which is insufficient for the remainder of the
proof
to gothrough.
More in-tuitively speaking,
hisproof
of Theorem C cannot beexpected
to work since it does notdistinguish
at all two facts related to theproblem :
(i)
the presence of the term - Ru in theequation
and(ii)
thecompactness
of 11/.Disregarding
thesefactors,
one would notegpect,
uniform convergence of asubsequence
ofthe Uq
but rathermerely
convergence in forany 8 >
0 to the trivialsolution,
as is the case with theproblem
where 92 is a bounded domain in Euclidean n-space. See
[4].
§
3 Partialresults.
We will show In the next section that a
subsequence
ofthe ug
con-verges in a certain sense to a smooth solution of
equation (2).
However the convergence is notstrong enough
toimply
thenon-triviality
of the resul-ting
solution. We demonstrate in this section that for alarge
class of ma-nifolds,
the convergence issufficiently
nice toguarantee
apositive,
smooth solution of(2).
THEOREM 2. There exists a
positive
constant 8(depending
ongii, R)
8itch that c, there exists cc
positive,
Coo solutionof equation (2)
withl~ = A. Thus Yamabe’s is true under this
assumption
on the metricgij.
PROOF. In the
equation (8)
consider a test functionThe result is
and hence from
(7)
Writing
tv =(u,)(9+1)11
the aboveinequality
becomesLet us
suppose A >
0 andapply
the Sobolev and Holderinequalities.
Weobtain,
thussince
II1tq
is boundedindependently
of q. Hence if forlarge enough q
and we obtainClearly
from(13),
thisinequality
continues to hold 0.Now
choose # N -
1. Then we obtainand
the uq
aresubsequently equibounded by
the Lemma. Asubsequence
therefore converges, with its
derivatives, uniformly
to a smooth solution of(2),
which is alsopositive by
the Lemma.271
Note that we have the
following
boundfor ),,
and therefore the conclusion of Theorem 2 will hold if
.lt1
cial case we have then
COROLLARY 1.
Any conzpact
C°° Riemannian1na.nifold
icithsealar can. be
defoi-med
to a C°° Riemannianof
constant scalar curvature.Corollary
1 may in fact be derived verysimply
from Theorem 1. Forimplies A C
0. Let the maximum value ot’ say be takenif
on at a
point
Then sinceJ(P)(0
we must haveand heuce
and
accordingly
the uq areequibounded.
Aubin
[1]
has shown that if ’It 2 3 and R is apositive constant,
then gij may betopologically
transformed into a metric gij with scalar curvature Rsatisfying
0. Hence as a furthercorollary
we haveCOROLLARY 2.
If
it >3,
a structureof’ positive,
constant scalfcr curva-tui-e niay be
deforiiied topologically
into oneof
constantnegative
8calar curva.Thus the
sign of
the tias notopological significane
inthan t1fO
4. A
regularity
Theorem.We prove now the
following
resultconcerning
weak solutions of’ equa- tion(
THEOREM 3. Let it be a
tt’’ (1t1)
solutionoj
aneqttatioit of
the(2).
Then 1i E Coo
(1lI ).
("S) The mean scalar onrvature is the quantity
PROOF. The function u satisfies
for
all ~ E j~~ {M ).
We choosean appropriate
testfunction ~ similarly
to amethod of J. Serrin
[5].
Defiue u = sup(u, 0)
and for afixed fl >
1(to
beeventually
chosen as in Theorem2)
ciefine the functionswhere
2q = p +
1.-
The function G
(1t)
is auniformly Lipshitz
continuous function of it and hencebelongs
to Likeuise Observe also that U and F vanish fornegative
u and thatLet us now substitute in
(17)
test functionswhere I
is anarbitrary, non-negative
C1 (M )
function. The resultis, using (7),
and hence
273
Using (19)
we then obtainLet us
take q
now to havecompact support
in a coordinatepatch
of M.The
integrals
in(21)
may then bereplaced by integrals
over asphere S
in En of radius R
where q
= r~(x)
ECo (R).
We choose R so thatThen
applying
the Holder and Sobolevinequalities
to(21)
we obtainand hence
We
choose fl
as in Theorem2,
ie 1C ~ ~ (n + 2)/(n
-2)
so that2q
N.Hence we may let in
(22)
to obtain the estimateLet
SR/2
denote thesphere
concentric to S of radiusR/2
and choose 1] = 1 on~2~ ! i ~~ ( c 2/R
onSR .
Then we obtainReplacing u by - u,
we obtain(24)
also for the function u andemploying
a
partition
ofunity clearly provides
aglobal
estimatefor some r
~> ~
where C will alsodepend
on the localLN
norms of ~c. The boundedness of it andsubsequently
its smoothness are now conseqnences ot’ the Lemma.Q.
F. D.For the functions uq defined in Section
2,
we have a uniform estimate for the normsand hence a
subsequence
converges in a weak solution of(2).
Hence we have
COROLLARY 3. 1’he
functions
v . inyV’_~ (ll~ ) J’or
any E>
0 toa > solution
of equation (2)
1c,ith A.Yamabe’s tlieorein would thus follow it’ the uon
triviality
of the above solution u could be demonstrated.This,
of course, we haveshown,
in thepreceding
section butonly
under a restriction A ~ ~ for certainpositive
c.Added in
proof (May 1968).
Since this paper waswritten,
’r. Aubin has found aproof
of Yamabe’s theoremby using
acompletely
different variation nalapproach.
REFERENCES
[1]
T. AUBIN. Sur la courbure scalaire des variétés riemanniennes compactes. C. R. Acad. Sci.Paris, Vol. 262, 1966 pp. A 130-133.
[2]
L. BERS, F. JOHN and M. SCHECHTER. Partial differential equations, Part. II. New York, 1964.[3]
J. MOSER. On Harnack’s theorem for elliptic differential equations, Comm. Pure Appl. Math., Vol. 14, 1961, pp. 577-591.[4]
S. I. POHOZAKV. Eigenfunctions of the equation 0394u + 03BBf(u) = 0, Dokl. Akad. Nauk.SSSR, Vol. 165, 1965 pp. 33-36.
[5] J.
SERRIN. Local behaviour of solutions of quasilinear equations, Acta Math. Vol. 113, 1965 pp. 219-240.[6]
S. L. SOBOLEV. Some applications of functional analysis in mathematical physics, Transl.Math. Monographs, Vol. 7, Amer. Math. Soc., Providence, R. I., 1963.
[7]
N. S. TRUDINGER. On Harnack type inequalities and their application to quasilinear ellip-tic equations. Comm. Pure Appl. Math., Vol. 20, 1967, pp. 721-747.
[8]
H. YAMADE. On a deformation of Riemannian structures on compact manifolds. Osaka Math.J., Vol. 12, 1960 pp. 21-37.